The relevance of IQ to the etiology and symptomatic expression of autism remains unclear. Although approximately three-quarters of individuals with autism have below-average IQ (< 70), autism can occur with equal severity in individuals of average and above-average IQ. While the finding that IQ can vary widely between identical twins with autism (Le Couteur et al., 1996
) would also suggest that it is not a useful marker of genetic subgroup differences in autism, there is nonetheless evidence from large-scale behavioral genetic studies (Bolton et al., 1994
; Fein et al., 1999
; Szatmari et al., 2000
) that IQ broadly defined (e.g., high versus low Verbal or Full Scale IQ) may index etiological heterogeneity and provide a basis for subtyping in autism.
In addition to ongoing attempts to elucidate the complex relationship of general intelligence, and particularly of mental retardation, to autism (Bailey, Phillips, & Rutter, 1996
), there has also been longstanding interest in the uneven intellectual abilities of individuals with autism. One question has been whether routinely administered IQ tests reveal any consistent pattern of cognitive strengths and weaknesses in autism. A Wechsler (1991
IQ profile with Verbal IQ (VIQ) depressed relative to Performance IQ (PIQ) and peak subtest score on Block Design (Happé, 1994
; Lincoln, Allen, & Kilman, 1995
) has traditionally been associated with autism and has even been suggested as a possible diagnostic aid (Lincoln, Courshesne, Kilman, Elmasian, & Allen, 1988). Although a recent meta-analytic review of 23 published studies (Lincoln, Courchesne, Allen, Hanson, & Ene, 1998
) confirmed that VIQ is generally lower than PIQ in autism, a VIQ < PIQ profile has not been found consistently across studies (see review by Siegel, Minshew, & Goldstein, 1996
). Some authors (Rumsey, 1992
; Siegel et al., 1996
) have attempted to explain these inconsistent findings by proposing that the difference between Verbal and Performance IQ in autism depends on the severity of impairment and diminishes as intellectual ability approaches the normal range. Yet, a VIQ < PIQ profile has been reported in several studies of high-functioning individuals with autism (see reviews by Lincoln et al., 1998
; Siegel et al., 1996
). It has also been suggested that the difference between Verbal and Performance IQ lessens with age and associated improvements in language functioning, at least among higher-functioning individuals with autism (Lincoln et al., 1998
). This hypothesis has been difficult to evaluate because most studies have included individuals spanning a wide range of ages, and very few studies have provided the opportunity to compare developmental differences between younger and older children.
VIQ–PIQ discrepancies have also played a prominent role in efforts to define Asperger syndrome as distinct from autism. Klin, Volkmar, Sparrow, Cicchetti, and Rourke (1995)
demonstrated a VIQ > PIQ discrepancy among a sample of individuals diagnosed with Asperger syndrome, opposite to the pattern typically found in autism. However, subsequent studies (Manjiova & Prior, 2000; Ozonoff, South, & Miller, 2000
) have failed to differentiate Asperger syndrome from autism on the basis of Wechsler IQ profiles. These inconsistent findings may reflect differences in the way Asperger syndrome was diagnosed. The Klin et al. study used the most stringent criteria, requiring a circumscribed interest and motor clumsiness for inclusion in their Asperger group. It is possible that these specific criteria selected for individuals with a VIQ > PIQ profile, as the development of circumscribed interests is likely to be mediated by verbal skills, and motor problems might be expected to be associated with deficits in PIQ. The problem of circularity in attempts to provide external validation for the distinction between autism and Asperger syndrome has been well recognized (e.g., Klin & Volkmar, 1997
; Volkmar & Klin, 2001
Questions of their broader nosological value aside, analyses of cognitive profiles within autism have suffered from two major limitations. First, emphasis on identifying a prototypical cognitive profile has been at the expense of considering the possibility of more than one cognitive profile in autism. At a conceptual level, recent advances in genetics and the understanding of autism as a complex disorder allow for the possibility of genetically meaningful variation in the cognitive phenotype of autism, which could potentially account for variation in the behavioral expression of autism. Yet, little attention has been paid to individual differences in cognitive profiles in autism. Particularly in the case of VIQ–PIQ differences, the report of group means for autistic samples could have the effect of ‘averaging out’ significant individual discrepancies that actually occur in both directions.
The second limitation concerns the usefulness of Wechsler subscale discrepancies for characterizing cognitive profiles in autism. Individuals with autism have frequently been reported to exhibit a characteristic pattern of unevenness in the subtest scores from which Wechsler Verbal and Performance IQs are derived (Happé, 1994
; Siegel et al., 1996
). For example, Block Design, a measure of visuospatial constructional ability, reliably yields the highest Wechsler subtest score in autism. In contrast, Picture Arrangement, another Performance subtest, but one which requires social-inductive reasoning, yields one of the lowest subtest scores on the Wechsler scales. Factor analysis has shown that Picture Arrangement and Block Design are weakly correlated and measure distinct abilities in individuals with autism (Lincoln et al., 1988
). Characteristic patterns of scatter are also found on the Wechsler Verbal scale. For example, Digit Span is regularly reported to be a relative strength on the Wechsler scales. Findings such as these raise questions about the interpretability of VIQ and PIQ scores in autism, and of any discrepancy that is (or is not) found between these scores.
In the present study, we addressed two main issues. The first was whether children with autism exhibit characteristic patterns of unevenness in their cognitive abilities and, if so, whether such patterns or profiles differ as a function of age or overall ability. To answer these questions, we assessed the cognitive abilities of a large sample of children with autism using the Differential Ability Scales (DAS; Elliott, 1991), a revised version of the British Ability Scales (Elliott, Murray, & Pearson, 1979
). The DAS is comprised of a Preschool and a School-Age battery. Each battery consists of only six core subtests, selected for their high loadings on psychometric g
, but with sufficient specific variance to support separate cluster scores. The DAS factor structure and corresponding cluster
scores are in accord with current models of human intelligence (Carroll, 1993
; Mackintosh, 1998
). They include Verbal
ability, the latter of which is further differentiated into Nonverbal Reasoning
ability in the School-Age battery. The DAS thus distinguishes between nonverbal subtests that primarily measure visual-spatial organization from nonverbal subtests that primarily measure inductive reasoning abilities.1
This is in contrast to the Wechsler system, which subsumes similarly divergent nonverbal subtests under its Performance subscale (Elliot, 1990).
The second issue was whether different cognitive profiles in autism are associated with differences in core symptomatology. To address this issue, we assessed children’s social and communicative functioning using the Autism Diagnostic Observation Schedule (ADOS; Lord, Rutter, DiLavore, & Risi, 1999
). The ADOS is a semi-structured, interactive observation scale that provides quantitative ratings of communicative and social behaviors corresponding to DSM-IV (APA, 1994
) and ICD-10 (WHO, 1993
) criteria for autism. The summary ratings from the ADOS allowed us to relate children’s neurocognitive profiles to quantitative measures of symptom severity in both the communication and social domains. This approach provided a way of potentially accounting for empirically defined phenotypic variance in autism while avoiding the methodological pitfalls involved in using a priori definitions, such as in recent attempts to differentiate Asperger syndrome from autism.